Creating A Boss AI for My Boss Rush Game

Before delving into the development of my boss gauntlet action game, I made the conscious decision to leverage premade art assets. As someone who doesn't possess artistic skills and lacks the time to acquire them, I find working with high-quality art assets incredibly motivating. Thus, I opted for the Penusbmic Dark Series Bundle 2, renowned for its stunning visuals. Interestingly, the artist behind this bundle has contributed to notable projects like Dome Keeper and Shogun Showdown, and their latest endeavor, Bullet Bunny, promises an exciting experience. 

Penusbmic's Page: https://penusbmic.itch.io

For my game's inaugural boss, I've selected the Blood King Pack, shaping our formidable adversary into a Bloody Swordsman.

Now, with a body in place, it's imperative to furnish it with a brain capable of dynamic operation. I'm not content with a boss that simply regurgitates a predefined set of actions; I aspire for it to respond intelligently to both player behavior and its own circumstances. Drawing inspiration from FromSoftware's Elden Ring and its boss Margrit the Fell Omen, I aim to replicate this nuanced approach to boss design.

To achieve this, I've turned to behavior trees to structure my AI. For those unfamiliar with behavior trees, Ai and Games offers an excellent lecture explaining their concept and application https://www.youtube.com/@AIandGames.

In essence, a behavior tree simplifies AI decision-making through a hierarchical, tree-like structure, with the root representing intent, branches depicting composite states, and leaves signifying resulting actions. Let's dissect the current tree.

At the outset, the selection state determines intent based on whether the last action executed was an attack. If not, it proceeds to select the appropriate attack, albeit randomly for now. Each attack comprises a sequence, adjusting its positioning based on the player's proximity.

I've strived for a balanced repertoire of ranged and melee attacks. These include a ranged attack that closes the distance, another that maintains distance, a melee attack with variable distance outcomes, and a melee attack that closes the distance.

Conversely, if the last action was an attack, the AI decides between a swift retreat to grant the player respite or a follow-up attack, contingent on the preceding attack and proximity upon its conclusion. Additionally, every follow-up attack resets the sequence, ensuring a quick retreat before reengaging.

With these decisions amalgamated, we've crafted a robust boss AI, albeit in its nascent stage. Presently, interaction is limited to maneuvering around the boss, as I've yet to implement player interaction. Consequently, the AI relies solely on positional data from both player and self.